Searching for New Physics using Classical and Quantum Machine Learning

Abstract

The development of machine learning (ML) has provided the High Energy Physics (HEP) community with new methods of analysing collider and Monte-Carlo generated data. As experiments are upgraded to generate an increasing number of events, classical techniques can be supplemented with ML to increase our ability to find signs of New Physics in the high-dimensional event data. This thesis presents three methods of performing supervised and unsupervised searches using novel ML methods.

The first depends on the use of an autoencoder to perform an unsupervised anomaly detection search. We demonstrate that this method allows you to carry out a data-driven, model-independent search for New Physics. Furthermore, we show that by extending the model with an adversary we can account for systematic errors that may arise from experiments. The second method develops a form of quantum machine learning to be applied to a supervised search. Using a variational quantum classifier (a neural network style model built from quantum information principles) we demonstrate a quantum advantage arises when compared to a classical network. Finally, we make use of the continuous-variable (CV) paradigm of quantum computing to build an unsupervised method of classifying events stored as graph data. Gaussian boson sampling provides an example of a quantum advantage unique to the CV method of quantum computing and allows our events to be used in an anomaly detector model built using the Q-means clustering algorithm.